Pairwise constrained concept factorization for data representation

Yangcheng He, Hongtao Lu, Lei Huang, Saining Xie

Research output: Contribution to journalArticlepeer-review


Concept factorization (CF) is a variant of non-negative matrix factorization (NMF). In CF, each concept is represented by a linear combination of data points, and each data point is represented by a linear combination of concepts. More specifically, each concept is represented by more than one data point with different weights, and each data point carries various weights called membership to represent their degrees belonging to that concept. However, CF is actually an unsupervised method without making use of prior information of the data. In this paper, we propose a novel semi-supervised concept factorization method, called Pairwise Constrained Concept Factorization (PCCF), which incorporates pairwise constraints into the CF framework. We expect that data points which have pairwise must-link constraints should have the same class label as much as possible, while data points with pairwise cannot-link constraints will have different class labels as much as possible. Due to the incorporation of the pairwise constraints, the learning quality of the CF has been significantly enhanced. Experimental results show the effectiveness of our proposed novel method in comparison to the state-of-the-art algorithms on several real world applications.

Original languageEnglish (US)
Pages (from-to)1-17
Number of pages17
JournalNeural Networks
StatePublished - Apr 2014


  • Clustering
  • Concept factorization
  • Data representation
  • Pairwise constraints

ASJC Scopus subject areas

  • Cognitive Neuroscience
  • Artificial Intelligence


Dive into the research topics of 'Pairwise constrained concept factorization for data representation'. Together they form a unique fingerprint.

Cite this